Machine Learning (Unsupervised)

[farsamo]

Hi

 

I doing unsupervised machine learning experiments. I used K-means algorithms however my question is how can I evaluate how good the data was grouped. I found something called Validity index.

 

Would you please guide me how to evaluate clusters.

 

Thanks

[elnetrino]

Try to use datasets where the documents are classifiied into known categories, i.e. datasets used for classification experiments. This is a nice way to examine the results of clustering algorithms. Use a k-means and initialize it with observations from each class. You can also specify number of clusters equal to the number of classes, and see if the clusters that are created match the classes.